如何计算矩阵的第一行与R中的每一行之间的余弦相似度？

Question

this is my_matrix :

ui 194635691 194153563 177382028 177382031 195129144 196972549 196258704   194907960 196950156 194139014 153444738 192982501 192891196
1 237      0.00      0.00      0.00      0.00      0.00      0.00         0      0.01         0         0         0         0         0
2 261      0.01      0.00      0.00      0.00      0.00      0.00         0      0.00         0         0         0         0         0
3 290      0.00      0.00      0.01      0.01      0.00      0.00         0      0.00         0         0         0         0         0
4 483      0.00      0.00      0.00      0.00      0.00      0.01         0      0.00         0         0         0         0         0
5 533      0.00      0.01      0.00      0.00      0.00      0.00         0      0.00         0         0         0         0         0
6 534      0.00      0.00      0.00      0.00      0.01      0.00         0      0.00         0         0         0         0         0

these are my codes are following:

b=my_matrix[1,2:length(my_matrix)]

for (i in nrow(my_matrix)) {
 res[i]=cosine(b,my_matrix[i,2:length(my_matrix)])
}

I used "lsa" package and I want to get a cosine similarity matrix that calculate b vector with every other vectors from matrix a but my codes throw a error that says :

argument mismatch. Either one matrix or two vectors needed as input.

What Should I do to fix my problem? many thanks in advance

Answer 1

Package "isa", which is not available for R version 3.2.2, is not really necessary. Just do it yourself, using the definition of cosine similarity:

my_matrix <- as.matrix(my_matrix)  # Make sure that "my_matrix" is indeed a "matrix".
v <- as.vector(my_matrix[1,-1])
M <- my_matrix[-1,-1]
cosSim <- ( M %*% v ) / sqrt( sum(v*v) * rowSums(M*M) )

The first line is only necessary if my_matrix is not yet a matrix but a data.frame .

A possible explanation for the original error message shown in the question:

I guess the class of the object my_matrix that was used in the code presented in the question and caused the error message

argument mismatch. Either one matrix or two vectors needed as input.

was data.frame , not a matrix . If so, the arguments b and my_matrix[i,2:length(my_matrix)] in the call of the cosine function are again data.frames, not a vector and a matrix as exspected.

As an aside:

Even if my_matrix is coerced to a matrix the code in the question will throw an error massage, since length(my_matrix) is larger than the number of columns and hence my_matrix[i,2:length(my_matrix)] selects undefined columns. The i -th row of my_matrix without the first column is my_matrix[i,2:ncol(my_matrix)] or shorter my_matrix[i,-1] .

Answer 2

you can try this:

A <- my_matrix[, -1]
b <- A[1,]
res <- apply(A[-1, ], 1, cosine, y=b)

This code was executed without an error:

d <- read.table(skip=1, text="ui 194635691 194153563 177382028 177382031 195129144 196972549 196258704   194907960 196950156 194139014 153444738 192982501 192891196
1 237      0.00      0.00      0.00      0.00      0.00      0.00         0      0.01         0         0         0         0         0
2 261      0.01      0.00      0.00      0.00      0.00      0.00         0      0.00         0         0         0         0         0
3 290      0.00      0.00      0.01      0.01      0.00      0.00         0      0.00         0         0         0         0         0
4 483      0.00      0.00      0.00      0.00      0.00      0.01         0      0.00         0         0         0         0         0
5 533      0.00      0.01      0.00      0.00      0.00      0.00         0      0.00         0         0         0         0         0
6 534      0.00      0.00      0.00      0.00      0.01      0.00         0      0.00         0         0         0         0         0")

my_matrix <- as.matrix(d)[,-1]  # without rownumbers.

library(lsa)
A <- my_matrix[, -1]  
b <- A[1,]
res <- apply(A[-1, ], 1, cosine, y=b)

But the result is vector with all values 0 (ie the first row is orthognal to the others). That depends on your data and is easily seen in this case.

Answer 3

The cosine function from the lsa package calculates the cosine measure between all column vectors of a matrix, therefore:

cosine(t(my_matrix[,2:ncol(my_matrix)]))

will return a matrix in which the first column is the vector of cosine measures between the first data row of my_matrix (b in your example) and all other rows.

If just want the vector of cosine similarities for the first row:

as.vector(cosine(t(my_matrix[,2:ncol(my_matrix)]))[,1])

The nth element of this vector is the cosine similarity between the first row and the nth row of the original matrix.

Answer 4

Let v be your 1 × m vector and M your m × n matrix

for (i in 1:dim(M)[2]){
  sim_cos_v[i] <- (v%*%as.vector(M[,i])) / (norm(as.matrix(v), "f")*norm(as.matrix(M[,i]), "f"))
}
sim_cos_v